Euler-Lagrange系统自适应模糊神经跟踪控制
发布时间：2018-08-03 19:25
【摘要】：Euler-Lagrange系统是一种具有代表性的非线性系统,它可以描述许多复杂的动力学问题,因此针对Euler-Lagrange系统轨迹跟踪控制的研究具有很好的实际应用意义和理论研究价值。本文提出了三种模糊神经自适应控制方法,为Euler-Lagrange系统的轨迹跟踪控制问题提供了有效的解决办法。首先,针对Euler-Lagrange系统中存在模型不确定性和未知外界扰动等问题,本文提出了一种基于变论域模糊系统的鲁棒自适应Backstepping跟踪控制方法。变论域模糊系统是由带有变伸缩因子的模糊基函数构成,通过伸缩因子根据系统状态的自适应在线调整,实现了模糊系统输入空间的自适应和模糊基函数的自适应,在不增加模糊规则的前提下提高控制精度。仿真研究验证了所提出方法的有效性。其次,为减少逼近器输入维度,降低运算复杂度,本文提出了一种基于极速学习神经网络的混合前馈-反馈鲁棒自适应跟踪控制方法。通过设计前馈极速学习神经网络逼近器,实现了对系统不确定性的有效逼近;与传统反馈逼近控制相比,所提出的前馈逼近器只需要参考量作为神经网络的输入,不仅减少了逼近器的输入维度,而且减少了隐含层节点数,从而极大精简了逼近器结构,降低了运算复杂度;此外,设计H∞鲁棒补偿项,消除未知外界扰动和逼近误差对控制精度的影响。仿真研究验证了所提出方法的有效性。最后,针对速度不可测的Euler-Lagrange系统,提出了一种基于自组织模糊神经观测器的H∞输出反馈控制方法。通过设计自组织模糊神经速度观测器,实现对未知速度的准确估计,并且该观测器能够自动在线生成模糊规则和修剪冗余规则,极大降低了运算复杂度;设计位置跟踪误差和速度误差相结合的滑模面,将系统不确定性和未知外界扰动重组为集总非线性,并设计自组织模糊神经网络逼近器对其进行在线自适应逼近;进而设计H∞鲁棒补偿项,进一步消除逼近误差,以提高控制精度和系统鲁棒性。仿真结果验证了上述方法的有效性。
[Abstract]:Euler-Lagrange system is a representative nonlinear system, which can describe many complex dynamic problems. Therefore, the research on trajectory tracking control of Euler-Lagrange system has a good practical significance and theoretical research value. In this paper, three kinds of fuzzy neural adaptive control methods are proposed, which provide an effective solution to the trajectory tracking control problem of Euler-Lagrange system. Firstly, a robust adaptive Backstepping tracking control method based on variable universe fuzzy systems is proposed to solve the problems of model uncertainty and unknown external disturbances in Euler-Lagrange systems. The variable domain fuzzy system is composed of fuzzy basis function with variable expansion factor. By adjusting the expansion factor according to the adaptive on-line state of the system, the adaptive input space of fuzzy system and the adaptation of fuzzy basis function are realized. The control accuracy is improved without adding fuzzy rules. Simulation results show that the proposed method is effective. Secondly, in order to reduce the input dimension of the approximator and reduce the computational complexity, a hybrid feedforward and feedback robust adaptive tracking control method based on extreme learning neural network is proposed in this paper. The feedforward learning neural network approximator is designed to achieve the effective approximation of the system uncertainty. Compared with the traditional feedback approximation control, the proposed feedforward approximator only needs reference as the input of the neural network. It not only reduces the input dimension of the approximator, but also reduces the number of hidden layer nodes, which greatly simplifies the structure of the approximator and reduces the computational complexity. In addition, the H 鈭，
本文编号：2162768
[Abstract]:Euler-Lagrange system is a representative nonlinear system, which can describe many complex dynamic problems. Therefore, the research on trajectory tracking control of Euler-Lagrange system has a good practical significance and theoretical research value. In this paper, three kinds of fuzzy neural adaptive control methods are proposed, which provide an effective solution to the trajectory tracking control problem of Euler-Lagrange system. Firstly, a robust adaptive Backstepping tracking control method based on variable universe fuzzy systems is proposed to solve the problems of model uncertainty and unknown external disturbances in Euler-Lagrange systems. The variable domain fuzzy system is composed of fuzzy basis function with variable expansion factor. By adjusting the expansion factor according to the adaptive on-line state of the system, the adaptive input space of fuzzy system and the adaptation of fuzzy basis function are realized. The control accuracy is improved without adding fuzzy rules. Simulation results show that the proposed method is effective. Secondly, in order to reduce the input dimension of the approximator and reduce the computational complexity, a hybrid feedforward and feedback robust adaptive tracking control method based on extreme learning neural network is proposed in this paper. The feedforward learning neural network approximator is designed to achieve the effective approximation of the system uncertainty. Compared with the traditional feedback approximation control, the proposed feedforward approximator only needs reference as the input of the neural network. It not only reduces the input dimension of the approximator, but also reduces the number of hidden layer nodes, which greatly simplifies the structure of the approximator and reduces the computational complexity. In addition, the H 鈭，
本文编号：2162768
本文链接：https://www.wllwen.com/kejilunwen/zidonghuakongzhilunwen/2162768.html
